AI Washing: When 'AI-Powered' Means Absolutely Nothing
Karani Geoffrey6 min readSomewhere between 2023 and today, the phrase "AI-powered" stopped meaning anything. It now sits on pitch decks, app-store listings, and pricing pages the way "natural" sits on a bag of crisps: technically undefined, legally slippery, and priced at a premium. My thesis is simple. AI washing is the deliberate practice of attaching the word "AI" to a product to inflate its perceived value and price, whether or not any meaningful artificial intelligence is involved. And once you learn to see it, you cannot unsee it.
What AI washing actually is
The term is a cousin of "greenwashing." Just as companies once slapped a leaf on a product to look eco-friendly, they now slap a neural-network graphic on it to look intelligent. The US Federal Trade Commission defines the problem in three flavours, and every one of them is worth memorising:
- The lie of existence: claiming a product uses AI when it is really if-statements, a lookup table, or a human quietly doing the work in the background.
- The lie of capability: the AI exists, but it cannot do what the marketing says. "Fully autonomous" turns out to mean "needs a human to check every output."
- The lie of hype exploitation: using AI buzz to lure buyers into schemes that were never going to work, AI or not.
In September 2024 the FTC launched Operation AI Comply specifically to prosecute these. Real cases followed: a company selling an "AI Lawyer" that could not practise law, a tool that used "AI" to generate fake customer reviews, and firms promising AI-run online storefronts that would make buyers rich. In 2025 the enforcement continued against a vendor, Air AI, over claims that its agents were autonomous enough to replace human employees. The through-line from the regulator: you must be able to substantiate every AI claim, explicit and implied. Most washers cannot.
Why this hits African businesses harder
In a market where a lot of buyers are new to the technology and the fear of "being left behind" is loudest, the AI label does more work. A local reseller can take a plain rules-based chatbot, rebrand it as an "AI customer intelligence platform," and charge triple, because the buyer has no easy way to check under the hood. Add the fact that compute and licensing already cost 25 to 40 percent more in Nairobi, Lagos, or Accra than in Europe, and AI washing becomes a direct tax on businesses that can least afford the waste.
The cruelty of AI washing in Africa is that it charges a premium for the word while quietly delivering the same old software, to buyers who are told that questioning the magic makes them unsophisticated.
How to see through the label in five minutes
You do not need a computer-science degree. You need to ask what the "AI" actually does and refuse to accept adjectives as answers. Here is the quick field test.
| The claim on the page | The question that exposes it |
|---|---|
| "AI-powered recommendations" | Does it learn from behaviour, or is it a fixed rule someone wrote once? |
| "Powered by advanced machine learning" | Trained on what data? Owned by you or the vendor? |
| "Autonomous AI agent" | What percentage of outputs need a human to approve them? |
| "Smart automation" | Is this AI, or is it just a well-configured workflow? (Often the latter, and that is fine, but do not pay AI prices for it.) |
| "Proprietary AI model" | Or is it a thin wrapper over a public model you could call yourself? |
The wrapper test
A huge share of "AI startups" are wrappers: a user interface bolted onto a public model like GPT or Claude, with a system prompt in the middle. There is nothing wrong with a wrapper if it solves your problem well. There is a lot wrong with paying a proprietary-model premium for one. Ask directly: "If I called the underlying model's API myself, how much of your product would I be rebuilding?" The honest ones will tell you their real value is the workflow, the integrations, the local support. The washers will get defensive.
The human-in-the-basement test
Some "AI" is people. It has happened repeatedly: services marketed as automated turned out to route work to human contractors offshore. Ask what happens at 2am on a public holiday when volume spikes. If latency and quality depend on a human being awake, the AI is not what you are buying.
When the AI label is fine, and when it is fraud
Let me be fair. Not every use of "AI-powered" is deception. If a company uses a genuine model to do genuine work and describes it accurately, the label is just a description. The line is substantiation. A legitimate vendor can show you the mechanism, the data flow, the accuracy numbers, and the limits. A washer offers only vibes, testimonials, and a countdown timer on the discount.
- Legitimate: "We fine-tuned a model on 40,000 of your industry's documents; here is our error rate and where it fails."
- Washing: "Our next-generation AI engine leverages cutting-edge intelligence to transform your business." (Notice: zero mechanism, zero numbers, all adjectives.)
The pricing tell
Watch what happens to the price when the AI label appears. A plain scheduling tool is one price. Add "AI-powered smart scheduling" and the same product often costs two or three times as much for a feature that is, functionally, the same calendar logic it always ran. This is the purest form of AI washing: the technology did not change, only the sticker did. In our market, where a genuine compute premium already inflates costs by 25 to 40 percent, paying an invented premium on top is a double loss. Before you accept any AI surcharge, ask the vendor to itemise exactly what the AI does that the cheaper version did not, and what that specific capability costs to run. A washer cannot itemise it, because there is nothing to itemise.
What buyers should actually do
- Ban the word in your evaluation. Ask the vendor to describe the product without ever saying "AI." If they cannot, there is nothing underneath.
- Price the non-AI version. Find out what a comparable non-AI tool costs. If the AI premium is large, demand to see the value that justifies it.
- Demand the mechanism in writing. A one-paragraph plain-language explanation of what the model does, on what data, hosted where. Washers hate paper trails.
- Check the claims against reality with a trial. The label survives the pitch. It rarely survives your own data.
The bottom line
"AI-powered" is not a feature. It is a marketing decision, and increasingly a legal liability for the companies that abuse it. Your defence is unglamorous and completely effective: ask what it does, ask for the mechanism, ask for numbers, and refuse to pay a premium for a word. Before you commit budget on the strength of a label, it is worth stepping back and asking whether your business is even ready to buy, a question I unpack in this piece on not getting pressured into AI before you are ready.
The companies building real AI want you to look under the hood, because that is where their advantage lives. The ones washing want you to admire the paint. Look under the hood every single time.